search

SamsungLabs / fcaf3d

[ECCV2022] FCAF3D: Fully Convolutional Anchor-Free 3D Object Detection
MIT License
229 stars 37 forks source link

Training on KITTI #12

Closed abhigoku10 closed 2 years ago

abhigoku10 commented 2 years ago

@filaPro hi thanks for opensourcing code base , does this architecture been trained on KITTI , ARGOVerse, Nuscense dataset? if not what changes have to be made for the training on those dataset Thanks in advance

filaPro commented 2 years ago

Hi @abhigoku10 ,

KITTI and nuScenes are supported in mmdetection3d, however all detectors there are either indoor or outdoor. You can check this issue. So basically it is not easy to apply FCAF3D for outdoor datasets. For example our detection head is multi-scale and sparse 3d, however all detection heads for KITTI are single scale and dense 2d.

abhigoku10 commented 2 years ago

@filaPro thanks for the response !!

zachluo commented 2 years ago

@filaPro hi, following the discussion, is there any work that has tried to apply voxelnet or pointpillar or others on the Scannet?

filaPro commented 2 years ago

Hi @zachluo , I am not aware of such works.